Labor Market Expectations and Major Choice forLow-Income, First-Generation College Students:

Evidence from an Information Experiment∗

Alexander I. Ruder† Michelle Van Noy‡

June 13, 2017

Abstract

We conduct a survey experiment with a large sample of U.S. college students toassess how labor market information affects earnings expectations and college majorchoice for students from different family backgrounds. We find labor market informa-tion about post-graduate earnings reduces students’ own earnings expectations, thoughmostly in students’ counterfactual majors. We find some evidence of treatment effectheterogeneity by SES. In several fields, high-SES respondents have higher expectationsthan low-SES respondents; the earnings treatment leads to lower earnings expectationsfor high-SES respondents, but has no substantive effect for low-SES respondents. Wefind no evidence that the information treatment changes the probability of choosinga major. Our results build on previous scholarship to suggest that labor market in-formation impacts earnings expectations though has little impact on choice relative toother factors in the student decision-making process.

∗Project funded by the Russel Sage Foundation project RSF #83-15-06.†Corresponding author: Assistant Professor, Department of Political Science, University of South Carolina

and Non-Resident Visiting Scholar, John H. Heldrich Center for Workforce Development, Rutgers University.Email: RuderA@mailbox.sc.edu. Phone: 1-803-777-3109. Postal address: Department of Political Science,University of South Carolina, 349 Gambrell Hall, 817 Henderson Street, Columbia, SC 29208‡Associate Director in the Education and Employment Research Center in the School of Management

and Labor Relations at Rutgers University, New Brunswick (mvannoy@rutgers.edu).

Despite large increases in overall college enrollment, there is continuing concern over so-

cioeconomic disparities in education. These disparities are particularly concentrated among

low-income students, minority students, and first-generation college students (Engle and

Tinto 2008). Low-income, first-generation college students have far lower baccalaureate de-

gree completion rates than their peers, and face many more barriers to success. As this

growing educational inequality in higher education is a threat to the nation’s economic

competitiveness, a variety of policy responses at national, state, and university levels are

targeting these groups.

This effort is focused not only on increasing attendance and completion, but also on

choosing a field of study (Davies and Guppy 1997; Goyette and Mullen 2006; Lundy-Wagner

et al. 2014; Ma 2009). Students from lower socioeconomic family backgrounds (SES) are

underrepresented in fields that are more likely to lead to high career earnings and job security,

such as science, technology, engineering, and mathematics (STEM). Among explanations for

why these discrepancies exist, scholars in both human and cultural capital theory traditions

suggest that lower-SES students lack the same degree of access to accurate information about

the costs and benefits of a higher education degree as their higher-SES peers (Coleman 1988;

Perna 2000; Perna and Titus 2005).

Our paper asks how labor market information influences student expectations about post-

graduate earnings in different college majors, and how this information relates to student

expectations about completing a degree in these different majors. We partner with a large,

public university system with three socioeconomically diverse campuses to conduct a survey

with an embedded information experiment. Previous studies with information interventions

typically focus on a smaller sample of students at a single college, while our survey includes

nearly 3000 students with diverse economic and social backgrounds and who study in different

institutional settings in the United States. We then examine whether low-income, first-

generation college students hold different earnings expectations than their higher-SES peers,

1

and how providing earnings information reduces these disparities and impacts choice.

To examine the impact of labor market information, we randomly assign respondents into

one of two conditions. One group sees no labor market information while the second group

sees median earnings of the U.S population of college graduates from four-year universities.

After seeing the information treatment, all respondents report—for each academic major

field—their expected future earnings and probability of completing a degree in each major

field.

Our main results are as follows. First, we find that students expect to earn above the

national median level of earnings, especially in the Business, Health, and STEM fields.

Showing respondents labor market information on median earnings reduces the amount they

expect to earn post-graduation towards the national median. Labor market information has

the largest effect on earnings expectations when respondents are asked about their future

earnings in alternative, or counterfactual, majors compared to the fields that they expect to

choose or have already chosen. This result is consistent with the informal model of major

choice proposed in Hastings et al. (2015), where the cost of acquiring information about

majors means that students are likely to have more knowledge about degree programs that

correspond to their interests.

Second, we estimate a series of regressions for degree completion probability by treatment

condition. We find no evidence that the information intervention affects students’ expected

probability of degree completion in different fields. Thus, our finding with respect to choice

is similar to scholarship that shows information treatments impact earnings expectations

though have little or no impact on educational choice (Kerr et al. 2014), suggesting factors

beyond earnings are the dominant determinants of choice (Hastings et al. 2017; Wiswall and

Zafar 2015).

Third, we find limited evidence that the information treatment differentialy affects stu-

dents from low-SES family backgrounds. We find low-SES respondents expect to earn less

2

than their peers in the Business, Education, and Humanities fields. The information treat-

ment reduces the discrepancy in Business and Humanities by lowering the earnings expec-

tations of high-SES respondents, rather than raising the earnings expectations of low-SES

respondents. In Education, the treatment reduces the SES-based discrepancy in earnings

expectations by increasing low-SES respondents’ earnings expectations. Finally, we find no

evidence that the information treatment effect varies by family background.

Our work contributes to research investigating the impact of earnings expectations on ed-

ucational choice (Baker et al. 2017; Hastings et al. 2015; Huntington-Klein 2016a,b; Hurwitz

and Smith 2016; Jensen 2010; Reuben et al. 2016). Our research adds to this literature by

examining the impact of a randomized information intervention on a large, diverse student

sample from a public university system in the United States. In addition, our partnership

with the university system gives us access to administrative data that includes a large number

of academic and demographic control variables.

Our work also builds on scholarship examining the relationship of socioeconomic discrep-

ancies in the knowledge of college labor market outcomes and educational choice (Beattie

2002; Betts 1996; McDonough and Calderone 2006; Walpole 2003). Similar to Bleemer and

Zafar (2015) and Hastings et al. (2015), we find lower-SES respondents have generally similar

post-graduate labor market expectations as their higher-SES peers.

Our approach of providing information builds on work that investigates the impact of

information interventions on educational choices (Bettinger et al. 2012; Fryer 2013; Hoxby

and Turner 2013; Kelly 2015; Nguyen 2013). Our between-group research design includes a

randomized information treatment, allowing us to assess the causal impact of median earn-

ings information on students’ earnings expectations and expected probability of completing

a degree in different fields.

The paper is organized in the following manner. In Section 1 we review the relevant

literature on labor market outcomes and student family background on education choice.

3

Sections 2 through 4 discuss the research questions, the survey and administrative data, and

the experimental design. Section 5 presents the results, and Section 6 concludes.

1 Major Choice and Family Background

1.1. Socioeconomic Status and Major Choice

An ongoing concern for policy researchers is understanding the persistent discrepancies in

higher education attainment and labor market outcomes for students from different socioe-

conomic groups. Scholars have documented SES discrepancies in college enrollment and per-

sistence (e.g., Beattie 2002; Engle and Tinto 2008) and major choice (e.g., Lundy-Wagner

et al. 2014; Ma 2009). Both economic theories of human capital and sociological theories of

cultural capital offer insight into these discrepancies by explaining how a student’s family

background can influence educational choice and outcomes.

Knowledge of the costs and benefits of college are not necessarily uniform across different

groups.1 Some scholars find that individuals from lower-income families make less accurate

estimates about the estimated benefits of a college degree (e.g., Betts 1996), and know less

about financial aid (Hoxby and Turner 2013; Olson and Rosenfeld 1984). However, Paulsen

(2001) concludes that students are “reasonably careful and accurate in their acquisition

of information about earnings differentials . . . [they] acquire information that is adequate

to make more or less economically rational college-going decisions” (Paulsen 2001, p. 63).

Similarly, Rouse (2004) finds that high- and low-income students have similar expectations

of the economic returns to college attendance. Hastings et al. (2015) and Bleemer and

Zafar (2015) find that low-income students hold less accurate beliefs about past graduates’

earnings, though students’ own beliefs are similar across income groups. Toutkoushian (2001)

shows that low-income, first-generation students expect to attend similar institutions as their

higher-SES peers.

1See Perna (2006) and Paulsen (2001) for reviews.

4

Many works have elucidated how family background impacts student career expectations

and employment outcomes. The advantages held by higher-SES students often begin before

students enroll at college and persist throughout college and into the labor force (Astin 1993;

Hu and Wolniak 2010, 2013; Lareau 1993; Walpole 2003). Lower-SES students form different

career expectations than their higher-SES peers (Ali et al. 2005; Metheny and McWhirter

2013; Trusty et al. 2000). Social and career-related barriers, such as lack of role models,

financial support, and internal and external barriers to career progress, influence career

outcome expectations (Lent et al. 2000). Similarly, Ma (2009) discusses the role of parental

involvement in the educational decision-making process and the domain-specific expertise

about careers that parents provide their children.

Betts (1996) offers several explanations for why family income is related to information

about the costs and benefits of college: higher income allows families to acquire more infor-

mation; working parents offer prospective college students an example of what employment

is like in specific careers; and information about the returns to college may not be extensively

disseminated in low-income neighborhoods.

Hastings et al. (2015) propose an informal model of degree choice that makes several

predictions related to a student’s socioeconomic status. In their model, students face uncer-

tainty over the earnings and costs of different degree programs. Students reduce uncertainty

by acquiring degree-specific information through a costly search process. Certain groups,

such as low-SES students, face higher search costs, as the process of acquiring this informa-

tion is likely more difficult. Their model predicts that low-SES students have less accurate

expectations about post-graduate earnings. They find support for this prediction when an-

alyzing a large sample of students in Chile, while Huntington-Klein (2016b) finds partial

support for the model’s prediction using a sample of U.S. high school students.

Sociological theories posit that students’ outlooks about majors and careers are inex-

tricably linked to their class-based experiences. Bourdieu’s work (Bourdieu 1984, 1986) on

5

cultural capital and habitus observes that higher-income families intergenerationally trans-

mit knowledge and dispositions that are valuable within institutions. In this way, infor-

mation about how to navigate educational institutions, as well as underlying expectations

about educational achievement, are inherent in students’ class-based experiences (Lareau and

Weininger 2003; McDonough 1997). Further, social capital includes an individual’s relation-

ship to social networks, which can connect to valuable resources or information. Access to

these social networks also varies by social class and can impact educational decision-making

(Coleman 1988; Perna 2000; Perna and Titus 2005).

The research builds on a growing body of literature that examines the link between post-

graduate earnings and educational choice. Early work focuses on how expected earnings

influence the demand for post-secondary education (Berger 1988; Manski and Wise 1983;

Willis and Rosen 1979). Building on Manski’s work (Dominitz and Manski 1996; Manski

1993; Manski and Wise 1983), more recent studies examine how students’ subjective ex-

pectations about future earnings influence major choice or college enrollment (Arcidiacono

et al. 2012; Attanasio and Kaufmann 2012; Delany et al. 2011; Jensen 2010; Wiswall and

Zafar 2015). This work suggests that student expectations of post-graduation earnings in-

fluence their educational choices, although the impact of preferences or taste is greater than

expected earnings.

All these works follow a human capital perspective of educational choice, where information—

knowledge of the costs, benefits, risk, and opportunities—plays a key role. An idealized hu-

man capital framework assumes that students possess perfect information about the costs,

benefits, and risk, and use that information to rationally choose the most efficient educa-

tional option. Scholars have investigated the validity of these assumptions by asking what

students actually know about post-graduate earnings outcomes, how expectations are mea-

sured, and how students expectations are formed in the first place (Avery and Kane 2004;

Beattie 2002; Betts 1996; Botelho and Pinto 2004; Manski 1993; Paulsen 2001).

6

1.2. Low-Income, First-Generation Students and Socioeconomic Status

Student socioeconomic status is a broad construct that includes a student’s financial, social,

cultural, and human capital resources (NCES 2012). Because SES includes so many different

factors of a student’s family and personal background, no consensus exists on how to mea-

sure it. In general, SES measurement involves an index of the “big three” items: parental

education, parental occupational status, and household or family income (NCES 2012).

Our focus in this project is on two components of the broader concept of student SES:

family educational status and family income.2 While only representing a partial component

of SES, low-income students who are the first in their family to attend college share many

of the same barriers to higher education as do most lower-SES students. In particular,

first-generation college students are likely to lack information sources about college and

different academic majors that are available to students with a college graduate in the family

(Horn and Nunez 2000; Ishitani 2006). Low-income students are likely to lack institutional

resources—such as quality school counselors and teachers—that help disseminate information

about college and careers (Avery 2009; Hoxby and Turner 2013).3

2 Research Questions

The human capital and social capital theoretical frameworks offer explanations for how in-

formation provision influences earnings expectations and educational choice. In addition, the

literature explains mechanisms for how family characteristics influence educational attain-

ment and earnings expectations: students from lower-SES families may lack equal access or

pay a higher cost to acquire information about the economic returns to different educational

choices.4 Motivated by each theoretical perspective’s focus on information and preferences

2Our survey includes no information about family occupational status.3At points in this manuscript, the terms “SES” and “family background” are used interchangeably for

stylistic reasons.4Paulsen (2001, p. 75) discusses the shared focus of these two theoretical approaches on the influence of

family background. Researchers who have used both theoretical models to study student educational choice

7

over majors, we study the following research questions:

1. Does providing students with labor market information about median earnings change

their own earnings expectations in different majors, relative to students who do not

see any labor market information?

2. Does providing students with labor market information about median earnings change

their own expectations about the field of study in which they intend to complete a

degree, relative to students who do not see any labor market information?

3. Does the effect of labor market information about median earnings vary systematically

by student’s socioeconomic status?

The first research question introduces the information treatment. Specifically, we ask

whether providing labor market information on the median earnings of different majors

changes beliefs about students’ future earnings. The second research question investigates

the impact of earnings information on the expected probability of completing degrees in

different majors. The third research question examines whether the effect of labor market

information varies by a student’s family background status. This final question assesses

whether information provision reduces any inequalities in information across different family

background groups. In particular, if low-income, first-generation students have biased esti-

mates of the population earnings in each major, then the information intervention should

have a larger effect on this group of students.

3 Data Source

Our data source is an original survey administered to the students at three separate campuses

of a large, public university system.5 To administer the survey, we partnered with the

include Ma (2009), Lundy-Wagner et al. (2014), and Wells and Lynch (2012).5We exclude respondents from one specialty campus that has its own admissions process and all students

must major in the field of healthcare. For these students, there is no option to change majors as that would

8

university system’s office of institutional research (OIR), which has extensive experience

fielding large surveys with the student body. Before OIR launched the survey, we designed a

university system-wide marketing strategy to raise awareness about the forthcoming survey.

Research assistants distributed fliers in high-traffic areas of campus, engaged with student

clubs and residential halls, and used social media accounts to post advertisements. We also

partnered with university staff in student affairs and other special offices whose primary

mission involves service to lower-SES students. Through these partnerships, we spoke to

several different student groups about the survey, or provided advertisement material to

administrative staff who then distributed it to the students via an email campaign and in-

person discussion. These outreach efforts began in September 2015 and continued until we

closed the survey.

OIR emailed the survey to all undergraduates at the three campuses on 11/3/2015. The

office sent follow-up emails on 11/9/2015, 11/16/2015, and 11/23/2015. The survey closed

on 11/30/2015.6 The email text explained the survey and the incentive structure. Students

who completed the survey were entered into a lottery for one of six $500 Visa gift cards

or one of ten $100 gift cards.7 The email included a link to an online survey hosted by

Qualtrics, where students agreed to take the survey by signing an online consent form.

The email invitation reached 48,139 undergraduate students. The response rate was 12.9

percent, with 6,243 students responding and 4,908 students completing the survey.8 As

stated above, we remove respondents from the health sciences campus, foreign students, and

imply leaving college. The results are not substantively different with this group of 164 students included.6Together with OIR, we decided to field the survey late in the fall academic semester, as OIR had another

survey in the field earlier in the semester and was concerned about survey fatigue. The office generally prefersto have only one survey in the field.

7Each week we randomly selected up to three winners from the list of completed respondents. With thewinners’ permission, we included their names and photos in the follow-up emails in order to encourage moreparticipation.

8We find no significant difference in observables between non-completers and completers. However, wefind some differences, common to online surveys, in those who choose or decline to respond to the survey.See supplementary materials for these statistics. We also remove foreign students from the sample.

9

98 respondents for which the administrative data is missing first-generation status.9

The full survey includes three randomized treatment conditions; we remove all respon-

dents in one of the three conditions, as the information provided in this condition is not

relevant to this current study.10 The remaining sample size is 2965 students.

The average time to complete the survey was 10 minutes and 53 seconds. Descriptive

statistics that compare the respondent sample to the overall university system and U.S.

undergraduate student population are presented in Table 1.

Sample University System NationalFreshman 22% 20% 25%Sophomore 19% 20% 19%Junior 27% 26% 21%Senior 32% 32% 28%Male 34% 48% 44%Caucasian 44% 40% 71%African American 11% 10% 16%Asian 24% 23% 6.8%Hispanic 16% 15% 12%Other 6% 5.5%SAT Math 605 603 522SAT Verbal 576 559 518First Gen. 20% 20% 31%Pell Grant 29% 28% 39%Business 17.5% 19% 20%Education 0.05% 0.06% 6.9%Health 8% 8.2% 12.2%Humanities 6.7% 5.9% 14%Other 6.6% 5.9 % 9.3%Social Science 13.1% 11.4% 18.6%STEM 17% 17% 17%Undeclared 31% 32% 1.9%

Table 1: Descriptive statistics of the sample and university data from the university office ofinstitutional research. U.S. data from the National Center of Education Statistics, 2008/2012Baccalaureate and Beyond (B&B). Large percentage of “undeclared” students in survey anduniversity data due to many students not officially declaring major until their third or fourthyear.

The descriptive statistics in Table 1 show that the sample is similar to the overall univer-

sity population on a number of observable demographic and academic characteristics. The

9The findings for the overall treatment effect are unchanged when including these cases. Though missingfirst-generation status, these cases have data on Pell status. We analyze the results using a Pell grantindicator only for SES status and find similar results.

10Since the treatment groups are formed through randomization, excluding one group results in no system-atic biases in the sample. The supplementary materials contains analysis showing that the randomizationprocedure successfully balanced observable characteristics across treatment groups.

10

one exception is gender, with the sample including a lower percentage of men than the overall

university population (34 percent in the sample versus 48 percent in the university system).

This gender discrepancy has been documented in the institutional research literature (see

Underwood et al. 2000).

We also compare the university system to the United States population of undergraduate

students in 2014. While similar on many dimensions, the university system has a different

racial profile, with a higher percentage of Asian students than the national student popula-

tion. The university system also has slightly higher SAT scores than the national average,

and a lower percentage of first-generation college students and Pell grant recipients. As such,

we use caution when extrapolating the results presented below to the general four-year col-

lege student population; however, the experimental design presented in the following section

allows us to measure, with a high degree of internal validity, the treatment effect on a large,

diverse convenience sample of college students.

3.1. Survey and Experimental Design

Upon agreeing to take the survey, respondents answer several questions about their educa-

tional background. We then ask questions related to their expectations of post-graduation

earnings and expected major choice. Specifically, students evaluate six major fields: Busi-

ness, Education, Healthcare (Health), Humanities, Social Science, and STEM.11 For each

major grouping, we ask students to consider the type of careers associated with each major,

and then to estimate their earnings if they were working full time in the fifth year after

graduation.12

11There are many specific majors in the university system. To reduce the number of options given tostudents, we group these individual majors into these categories, which are similar to those used, for example,in Caner and Okten (2010). We define STEM as a combination of engineering, biological and physicalsciences, and computer science.

12We ask students to consider five years for two primary reasons: 1) Asking students to consider a periodright after graduation would perhaps not allow for sufficient time to find a job in the major; and 2) We allowstudents who plan to attend graduate school time to complete their advanced degrees and enter employment.

11

Earnings question: If you were to receive a Bachelor’s degree in each of the

following fields of study and you were working full time 5 years after graduation,

what do you believe is the most likely amount that you would earn per year?

Major Choice: What is your likelihood of completing a degree in each field of

study? That is, what do you believe is the percent chance (or chances out of a

100) that you would graduate with a major in each of the following categories?

The first question asks respondents to estimate their future earnings outcomes in the six

major groups. Respondents use a slider to pick the dollar amount they expect to earn in

each major. We then ask respondents to report their expected probability of completing a

degree in each of the major groups.13 Respondents type the numbers directly into six cells

that together must sum to 100.14

3.2. Treatment Description

The following two experimental treatments allow us to answer whether information provision

changes students’ estimates of labor market outcomes and major choice, relative to a baseline

where students see no labor market information:15

No Information: Respondents receive no labor market information.

Median Earnings: Respondents receive the median earnings of graduates in each major

field.

13The wording for both these questions is similar to Wiswall and Zafar (2015).14As in Baker et al. (2017) and Wiswall and Zafar (2015), the probabilities are constrained to sum to one

in order to approximate a real-world choice situation. For some particular research questions, this constraintalso allows the researcher to use a generalized ordinal logit model to estimate the choice probabilities

15Data source for labor market information is the U.S. Department of Education, National Center for Ed-ucation Statistics, 2008/12 Baccalaureate and Beyond Longitudinal Study (B&B), accessed through NCESPowerStats. We use data from the year 2012 so that data reflects respondent status four years after grad-uation. Respondents were not told the data are from 2012. To correspond with our question wording,respondents were told that the data was from five years after graduation rather than four. We limit B&Bdata to graduates who are working full-time. See supplementary materials for the survey instrument andthe tabular and graphical data displays we show respondents.

12

3.3. Pretesting

We pretested the survey instrument and experimental design for several months before the

November launch. In total, we had 70 test responses spread over 9 different test surveys.

We worked with several administrative offices in the university system in order to ensure the

pretesting subjects represented the economic diversity of the students in the university sys-

tem. These surveys tested different branching schemes, question types, and digital layouts.

The first tests took place in August and September 2015. As the surveys were tested, we

revised question wording, graphics, and layouts. The tests continued into October 2015, and

the survey was finalized for distribution the first week of November 2015.

3.4. Administrative Data

Through our partnership with the university system’s office of institutional research, we use

unique student identifiers to match survey responses to an administrative database of student

and family characteristics. These data include demographics and academic information, as

well as our key variables of student first-generation status and our measure of family income,

which is whether or not the student receives a Pell grant.16

4 Empirical Strategy

The first and second research questions are causal: we ask whether labor market information

affects expectations of post-graduate earnings and expectations of major completion. Though

treatments are randomized, the third question is descriptive in nature. We cannot randomly

assign family background to respondents, but we can assess evidence for treatment effect

heterogeneity by first-generation and low-income status. Similar to Engle and Tinto (2008),

we divide family background into three mutually exclusive groups. We identify whether

16The university office specifically identifies each student as first generation or not. The data on familyincome dollar amount are missing in too many instances to be of use; we therefore use Pell grant status asthe alternative.

13

a respondent’s family background is low-income and first-generation, low-income or first-

generation, or does not belong to either of these groups. With this coding scheme, our intent

is to create variation in socioeconomic status such that the low-income or first-generation

group has one socioeconomic disadvantage, while the low-income and first-generation group

has two socioeconomic disadvantages.

We observe several outcome variables in our survey. The first outcome variable is the

annual dollar amount a respondent expects to earn post-graduation for each major field; the

second outcome is the expected probability of completing a degree in each major field. Each

outcome measure requires a different estimation strategy.17

4.1. Students’ Preferred Major versus Counterfactual Major

Our analysis asks respondents about their labor market expectations for their own major

and counterfactual majors that they are not currently pursuing or interested in pursuing.

Respondents may have more knowledge about the labor market outcomes for their own

preferred major since they have a greater incentive to seek out information for the types of

careers they are interested in pursuing than for those in which they lack interest (Arcidiacono

et al. 2012). Costly search may also reduce student effort to learn about majors outside of

their interest area (Hastings et al. 2015).

4.2. Estimation of Earnings Expectations

To measure the relationship of family background and the median earnings treatment on

earnings expectations, we use ordinary least squares to estimate the following equation:

ln wik = β0k + β1k ∗ Ti1 + β2k ∗ FBia + β3k ∗ FBib + β4kXi + εi,k, (1)

17We top-code earnings responses at $150,000 to reduce the influence of any outlier responses. We alsocalculate the probabilities from the percentages, which we then recode from 0 to 0.001 and from 1 to 0.999.This allows us to take the log of the probability, and, similar to Wiswall and Zafar (2015) and Baker et al.(2017), allows for the use of a log-odds model of choice, which we use in a related paper.

14

where ln wik is the log wages of respondent i in major k, T1 is an indicator for the Median

Earnings treatment, FBa is an indicator for family background status that is first generation

or low income, and FBb is an indicator for family background status that is first generation

and low income. The vector Xi includes the following individual-level covariates: gender,

age, race, grade point average, ordinal ranking of probability of attending graduate school,

campus, and class level in school (first-year through senior).

To answer the first research question, we focus on the coefficient estimates of β1 and

test whether the estimates are substantively meaningful and statistically different from zero.

To answer the third research question, we focus on the estimated interactions between the

treatment and family background indicators.

4.3. Estimation of Major Completion

Our second outcome measure is the log probability of choosing and completing a degree in

major group k. We use ordinary least squares to estimate the following equation:

ln πik = β0k + β1k ∗ Ti1 + β2k ∗ FBia + β3k ∗ FBib + β4kXi + εi,k, (2)

where ln πik is the log expected probability of completing a degree in group k. The covariates

are the same as in Equation 1. Our specification to estimate major completion probability

is similar to Baker et al. (2017).

For the choice analysis, we limit our analysis to first-year, sophomore, and junior students

only, who are likely to face fewer academic and financial barriers to switching majors than

senior students. Examples of barriers to switching includes the investment in prerequisite

coursework for the current major. Switching majors may require additional coursework and

effort, delaying graduation and increasing the overall costs of college attendance. Results

obtained using the full sample of students are substantively similar.

In the choice analysis, 246 respondents answer by assigning the same maximum proba-

15

bility of completing a major to more than one major (for example, 50% for Social Science

and 50% for Humanities). Of these 246 respondents, 200 select two majors to be their most

likely major. These respondents are likely dual majors, while other respondents are likely

undecided. Results reported below are substantively similar if we include a dummy variable

for these respondents or omit them from the analysis.

5 Results

We now present the estimation results for Equations 1 and 2. We first present the results for

the impact of the information treatment on earnings expectations. Second, we show estimates

of the impact of the information treatment on major completion probability. Finally, we

present our estimates of heterogeneous treatment effects by family background status.

5.1. Earnings Results

We find strong evidence that the Median Earnings information treatment affects earnings

expectations. The results from Equation 1 are presented in Table 2. Each column contains

separate regression results for the six major fields. Heteroskedasticity-robust standard errors

are in parentheses. To evaluate the statistical significance of the estimates, we mark coef-

ficients with stars that represent significance at the 0.1 level, 0.05 level, the 0.01 level, and

a more stringent 0.0042 level. The 0.0042 level is a conservative significance level obtained

using the Bonferroni method to account for testing multiple hypotheses.18 All the regressions

in Table 2 include the covariates described in Equation 1. For presentation reasons, we omit

several covariates from the tables to focus on the treatment indicators. We present the full

tables in the supplementary materials.19

18We adjust for the α to account for 12 hypothesis tests in Table 2: 6 separate regressions and 2 dependentvariables (earnings and major choice).

19Regression specifications we present feature low R2, which suggest the explanatory variables and treat-ment explain little variation in the dependent variables. A concern with such low R2 values is omitted variablebias, but the treatment estimates are still unbiased due to the randomization procedure (Wooldridge 2012,p. 201).

16

Business Education Health Humanities Social Science STEMMedian Earnings −0.10∗∗∗ −0.02 −0.08∗∗∗ −0.06∗∗∗ −0.06∗∗∗ −0.08∗∗∗

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)

Low Income or First Gen. −0.07∗∗∗ −0.03 −0.03 −0.02 −0.01 −0.03(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)

Low Income and First Gen. −0.08∗∗ −0.04 −0.04 −0.04 −0.05 0.01(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)

Gender 0.06∗∗∗ 0.09∗∗∗ 0.03· 0.05∗∗ 0.02 0.03(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)

Age 0.00 0.01∗∗∗ −0.00 0.01∗∗∗ 0.01∗∗∗ −0.00(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)

R2 0.04 0.02 0.02 0.02 0.03 0.02N 2957 2957 2957 2957 2957 2957∗∗∗p < 0.0042, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table 2: OLS estimates of expected earnings per major on treatment indicator. Regressions alsoinclude family background indicators, and demographic and academic controls. Robust standarderrors in parentheses. Three stars indicate statistical significance at the level determined by theBonferroni correction.

In the row Median Earnings of Table 2, all the coefficient estimates are negative and

five out of six reach levels of statistical significance surpassing even the Bonferroni-corrected

critical value. The results also have substantive significance. For the Business column, the

estimate on the Median Earnings indicator of -0.10 results in a -9.5 percent decrease in log

expected earnings. The -0.08 estimate on the Median Earnings indicator in columns Health

and STEM results in a -7.7 percent decrease in log expected earnings. The percent changes

for columns Humanities and Social Science are smaller, equaling a -5.8 percent change in log

expected earnings.

Tables 3 and 4 suggest that earnings information has the greatest effect on respondents

who are asked about earnings in a field other than their own preferred major. Table 3

shows results for earnings expectations about respondents’ preferred major. In this table,

the Median Earnings indicator across major fields has smaller effects in magnitude and

less precise estimates than the pooled estimates in Table 2. Only among those who prefer

Business majors is the Median Earnings indicator statistically significant at the Bonferroni-

17

corrected value.

Business Education Health Humanities Social Science STEM

Median Earnings −0.09∗∗∗ −0.04 −0.05 0.08 −0.03 −0.04·

(0.03) (0.06) (0.04) (0.06) (0.04) (0.02)

Low Income or First Gen. −0.05 −0.10 −0.01 0.03 −0.03 −0.04(0.03) (0.07) (0.05) (0.07) (0.04) (0.03)

Low Income and First Gen. −0.11∗ −0.29∗ −0.00 0.05 −0.04 −0.03(0.05) (0.11) (0.06) (0.08) (0.06) (0.04)

Gender 0.07∗∗ 0.07 0.11∗ 0.06 −0.03 0.07∗∗∗

(0.03) (0.08) (0.05) (0.06) (0.05) (0.02)Age −0.00 0.01 0.00 0.00 0.01 −0.01

(0.00) (0.01) (0.00) (0.00) (0.00) (0.01)

R2 0.06 0.12 0.04 0.05 0.08 0.04N 739 252 486 326 544 1073∗∗∗p < 0.0042, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table 3: OLS estimates of expected earnings per major on treatment indicator. Regressions alsoinclude family background indicators, and demographic and academic controls. Sample limitedto responses about preferred major. Robust standard errors in parentheses. Three stars indicatestatistical significance at the level determined by the Bonferroni method correction.

Business Education Health Humanities Social Science STEM

Median Earnings −0.10∗∗∗ −0.02 −0.09∗∗∗ −0.08∗∗∗ −0.06∗∗∗ −0.10∗∗∗

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)

Low Income or First Gen. −0.08∗∗∗ −0.02 −0.03 −0.03 −0.02 −0.03(0.02) (0.02) (0.02) (0.02) (0.02) (0.03)

Low Income and First Gen. −0.06· −0.01 −0.06 −0.05 −0.06 0.03(0.03) (0.03) (0.04) (0.04) (0.04) (0.04)

Gender 0.04· 0.09∗∗∗ 0.04∗ 0.06∗∗ 0.04∗ −0.02(0.02) (0.02) (0.02) (0.02) (0.02) (0.03)

Age 0.00 0.01∗∗∗ 0.00 0.01∗∗∗ 0.01∗∗∗ 0.00(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)

R2 0.04 0.02 0.02 0.03 0.03 0.02N 2218 2705 2471 2631 2413 1884∗∗∗p < 0.0042, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table 4: OLS estimates of expected earnings per major on treatment indicator. Regressions alsoinclude family background indicators, and demographic and academic controls. Sample limited toresponses about counterfactual majors. Robust standard errors in parentheses. Three stars indicatestatistical significance at the level determined by the Bonferroni correction.

18

In contrast, Table 4 shows results when we elicit expectations about respondents’ coun-

terfactual majors. This table reveals that the information treatment has the largest effects

on respondents when asked about earnings in a major other than their own. The Median

Earnings treatment indicator is substantively large and statistically significant in all major

fields except Education.20

5.2. Choice

In Table 5, we present the OLS estimates of Equation 2. While the magnitude of the Business

and Education coefficient estimates in row Median Earnings are substantively large, none

of the estimates reaches conventional levels of statistical significance. All estimates are

imprecise with relatively large standard errors. Thus, we cannot reject the null hypothesis

of zero effect.

Business Education Health Humanities Social Science STEM

Median Earnings −0.08 0.11 0.03 −0.03 0.01 0.06(0.13) (0.11) (0.12) (0.11) (0.12) (0.13)

Low Income or First Gen. −0.06 0.07 0.17 0.04 0.28· 0.11(0.15) (0.13) (0.14) (0.13) (0.14) (0.15)

Low Income and First Gen. −0.21 0.19 0.20 0.26 0.14 0.06(0.23) (0.21) (0.23) (0.22) (0.22) (0.24)

Gender 0.43∗∗∗ −0.27∗ −0.57∗∗∗ −0.37∗∗∗ −0.47∗∗∗ 0.88∗∗∗

(0.14) (0.12) (0.12) (0.12) (0.13) (0.13)Age −0.01 0.00 −0.04∗∗ 0.03 0.03 −0.05∗∗∗

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)

R2 0.04 0.02 0.06 0.03 0.04 0.08N 2005 2005 2005 2005 2005 2005∗∗∗p < 0.005, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table 5: OLS estimates of expected probability of completion per major on treatment in-dicator. Sample excludes senior respondents. Regressions also include family backgroundindicators, and demographic and academic controls. Robust standard errors in parenthe-ses. Three stars indicate statistical significance at the level determined by the Bonferronicorrection.

20See the supplementary materials for details on coding respondents “in major.”

19

5.3. Family Background and Information Treatments

In Section 1, we discuss several reasons why low-income, first-generation students may hold

different labor market expectations than their higher-SES peers. Ma (2009) describes how

the parents of higher-SES students are more involved in educational decision-making, while

Hastings et al. (2015) note that low-SES students have higher search costs than their peers

since labor market information is not as readily accessible to them. Before reporting earnings

expectations and choice, we first examine whether low-SES respondents are more likely to

report low parental involvement in their educational decision-making process. Low parental

involvement can lead to higher search costs for major-specific information.

Our survey contains several questions that allow us to gauge whether low-SES respondents

have less parental involvement in educational decision-making than their higher-SES peers.

Table 6 shows descriptive statistics for answers to the question, “I can rely on my family

to:” followed by the prompts: help advise me on my selection of college major; help advise

me on my selection of career; advise me on how to find a job or internship; introduce me

to people who can help me get a job or internship; provide me with financial support while I

am getting my career started. For each prompt, respondents reply on an ordinal scale from

“never” to “all of the time.” For simplicity of presentation, we do not present the percentages

for the full ordinal scale. Rather, we calculate the share of respondents answering “never”

or “rarely” for each prompt.21

The descriptive statistics in Table 6 show that low-SES students are more likely to report

being unable to rely on parents for education and career-related decision-making. For all

prompts, low-SES students are more likely to report being able to “never” or “rarely” rely

on their parents. All the differences are statistically significant. To the extent that the lack

of parental involvement raises search costs for low-SES students, these differences can help

21Alternatively, we could calculate the share of respondents answering “often” or “all of the time.” Theresults are substantively similar.

20

explain heterogeneity of earnings expectations by family background status.

Parental Aid NeitherLow-Income or

First Gen.Low-Income and

First Gen.Advise major 0.19 0.34∗∗ 0.43∗∗

Advise career 0.17 0.31∗∗ 0.41∗∗

Advise finding job/internship 0.24 0.42∗∗ 0.56∗∗

Career networking 0.30 0.44∗∗ 0.59∗∗

Financial support during job search 0.12 0.26∗∗ 0.33∗∗

∗∗p < 0.01

Table 6: Reliance on parents for career-related guidance by family background status. Tableshows average of dichotomous variable equal to one if respondents state they “never” or“rarely” can rely on their parents for the given activity, and zero otherwise. Significance testsfrom linear regression of dichotomous dependent variable on indicators for family backgroundstatus.

5.3.1 Earnings Expectations

Descriptive statistics of log expected earnings by family background status are shown in Ta-

ble 7. Respondents from low-income and first-generation backgrounds have similar earnings

expectations when compared to their higher-SES peers. The difference in means is statis-

tically significant only in Business and Education, where low-income and first-generation

students expect to earn less than their higher-SES peers.

21

Major Family Background MeanBusiness Neither 11.25

Low-Income or First Gen. 11.18∗∗

Low-Income and First Gen. 11.14∗∗

Education Neither 10.74Low-Income or First Gen. 10.72Low-Income and First Gen. 10.67∗∗

Health Neither 11.17Low-Income or First Gen. 11.17Low-Income and First Gen. 11.15

Humanities Neither 10.67Low-Income or First Gen. 10.64Low-Income and First Gen. 10.60

Social Science Neither 10.74Low-Income or First Gen. 10.76Low-Income and First Gen. 10.72

STEM Neither 11.33Low-Income or First Gen. 11.31Low-Income and First Gen. 11.32

∗ ∗ p < 0.05, ∗p < 0.1

Table 7: Mean expected earnings by family background status. P-values from two-sideddifference in means test shown. The Low-Income or First-Generation group and the Low-Income and First-Generation group are separately compared to the base group that is nei-ther low-income nor first-generation. Category “Neither” refers to neither low-income norfirst-generation respondents. Statistics calculated using only respondents in the baselinecondition.

Table 8 shows the regressions for log earnings on the treatment indicators, the family

background indicators, and their interaction terms. The covariates are the same as the

specifications in Table 2. We find little evidence for heterogeneous effects by family back-

ground status. Only the Education interaction estimate is statistically distinguishable from

zero at the 0.05 level. The positive and substantively large coefficient on the interaction

term Median Earnings x Low-Income and First-Gen. suggests that median earnings infor-

mation increases the expectations of low-income and first-generation respondents compared

to similar respondents who see no information.22

22Similar to the pooled sample result, we find that the treatment impact is largest when respondents

22

We also use a F-test to test the null hypothesis that the interaction terms between

treatment indicator and family background are jointly equal to zero. Only one of the six

F-tests, for Education, achieves statistical significance.

Business Education Health Humanities Social Science STEMMedian Earnings −0.11∗∗∗ −0.04∗ −0.07∗∗∗ −0.08∗∗∗ −0.07∗∗∗ −0.08∗∗∗

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)Low Income or First Gen. −0.07∗ −0.04 −0.01 −0.05 −0.02 −0.02

(0.03) (0.03) (0.03) (0.03) (0.03) (0.02)Low Income and First Gen. −0.12∗∗∗ −0.10∗ −0.03 −0.09∗ −0.06 −0.01

(0.04) (0.04) (0.05) (0.04) (0.04) (0.04)Gender 0.06∗∗∗ 0.09∗∗∗ 0.03· 0.06∗∗ 0.02 0.03

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)Age 0.00 0.01∗∗∗ −0.00 0.01∗∗∗ 0.01∗∗∗ −0.00

(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)Median Earnings x Low Income or First Gen. −0.01 0.03 −0.04 0.05 0.02 −0.01

(0.04) (0.04) (0.04) (0.04) (0.04) (0.04)Median Earnings x Low Income and First Gen. 0.09 0.12∗ −0.04 0.11· 0.02 0.05

(0.06) (0.06) (0.06) (0.06) (0.06) (0.05)R2 0.04 0.02 0.02 0.02 0.03 0.02F-Test p-value 0.22 0.10 0.59 0.17 0.90 0.55N 2957 2957 2957 2957 2957 2957∗∗∗p < 0.0042, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table 8: OLS estimates of expected earnings per major on treatment indicator, familybackground indicators, interactions between treatment and family background indicators,and demographic and academic controls. Robust standard errors in parentheses. Three starsindicate statistical significance at the level determined by the Bonferroni method correction.

Taken together, we find that for the major fields in which earnings discrepancies exist by

family background, the information treatment tends to reduce the discrepancies. In Table 8,

the estimates in row Low Income and First Gen. show that low-income and first-generation

respondents expect to earn less than their high-SES peers in Business, Education, and Hu-

manities. The estimates for these major groups are substantively large and statistically

significant at least at the 0.05 level. However, in the Median Earnings group, these differ-

ences are reduced, as indicated by the positive interaction terms in row Median Earnings x

Low Income and First Gen.

To better illustrate this result, we use the regression estimates to predict the average

report earnings expectations of counterfactual majors. These regressions are presented in the supplementarymaterials.

23

expected earnings by treatment group and family background status.23 Table 9 shows the

predictions.24 For Business, Education, and Humanities, we find a statistically significant

difference in expected earnings between family background groups in the No Information

condition, while in the Median Earnings condition we find that the difference is statistically

indistinguishable from zero. For Business majors in the No Information condition, the es-

timated difference in earnings between family background groups is 0.12. In contrast, for

Business majors in the Median Earnings group, the estimated difference is 0.02 and the 95%

confidence interval overlaps zero. The difference is reduced due to the treatment reducing

high-SES respondents’ earnings expectations more than low-SES respondents’ expectations.

Respondents in the Humanities major group present a similar result. The predicted

difference between family background groups in the No Information condition is 0.09, while

the predicted difference in the Median Earnings condition is -0.02 with a 95% confidence

interval that overlaps zero. As with the Business major group, the difference is reduced due

to the treatment reducing high-SES respondents’ earnings expectations more than low-SES

respondents’ expectations.

23We calculate the expected value of the dependent variable, as well as the estimated difference and 95%confidence interval, using the simulation function of the Zelig (Imai et al. 2008) package for the statisticalsoftware R.

24Formally, the estimated difference is defined as ∆ = E[π|Median Earnings, X]−E[π|No Information, X],where ∆ is the difference, π is the probability of completing degree k, and X is a vector of covariates.

24

Predicted EarningsMajor/Treatment Condition Low-SES High-SES Difference CIBusiness (No Information) 11.13 11.25 0.12 (0.04, 0.2)Business (Median Earnings) 11.11 11.14 0.02 (-0.05, 0.12)Education (No Information) 10.6 10.7 0.1 (0.02, 0.17)Education (Median Earnings) 10.68 10.65 -0.03 (-0.11, 0.37)Health (No Information) 11.10 11.13 0.03 (-0.05, 0.12)Health (Median Earnings) 11.00 11.06 0.06 (-0.02, 0.15)Humanities (No Information) 10.53 10.62 0.09 (0.00, 0.18)Humanities (Median Earnings) 10.55 10.54 -0.02 (-0.11, 0.08)Social Science (No Information) 10.64 10.70 0.06 (-0.03, 0.14)Social Science (Median Earnings) 10.59 10.63 0.04 (-0.05, 0.12)STEM (No Information) 11.31 11.32 0.02 (-0.06, 0.09)STEM (Median Earnings) 11.27 11.24 -0.03 (-0.11, 0.05)

Table 9: Predicted mean expected earnings (log) in preferred major, by treatment conditionsand family background status. Column Low-SES shows predicted mean expected earnings forlow-income and first-generation students; column High-SES shows predicted mean expectedearnings for neither low-income nor first-generation students. Column Difference showsthe estimated difference in expected earnings between the two SES groups. Column CIreports the 95% confidence interval on the estimated difference. Predictions calculated fromregression estimates in Table 8.

Unlike Business and Humanities, the difference for Education is reduced due to the

treatment increasing low-SES respondents’ earnings expectations more than high-SES re-

spondents’ expectations. The predicted difference between family background groups in

the No Information condition is 0.1, while the predicted difference in the Median Earnings

condition is -0.03 with a 95% confidence interval that overlaps zero.

5.3.2 Choice

We find no evidence for heterogeneous effects of the earnings treatment on major completion

probability. Table 10 shows results from the effect of the earnings treatment and interac-

tions with family background on expected major of completion. Conditional on the model

covariates, respondents from different family backgrounds have similar major completion

probabilities. In the No Information condition, we find no statistically significant difference

25

in major completion probability between low-income and first-generation respondents and

their higher-SES peers. However, three estimates in the row Low Income and First Gen. are

positive and substantively large: the estimate is 0.48 for Education, 0.50 for Humanities,

and 0.42 for Social Science, suggesting low-SES respondents are more likely to select these

majors. The estimates, however, are not statistically significant at the 0.05 level. The esti-

mates for Business, Health, and STEM in the row Low Income and First Gen. are closer to

zero and statistically imprecise.

Business Education Health Humanities Social Science STEMMedian Earnings −0.04 0.22 0.02 0.04 0.08 0.05

(0.16) (0.14) (0.15) (0.14) (0.15) (0.16)Low Income or First Gen. −0.04 0.16 0.14 0.09 0.32 0.12

(0.21) (0.19) (0.20) (0.18) (0.20) (0.21)Low Income and First Gen. −0.07 0.48· 0.23 0.50 0.42 −0.01

(0.32) (0.29) (0.31) (0.31) (0.32) (0.33)Gender 0.43∗∗∗ −0.27∗ −0.57∗∗∗ −0.38∗∗∗ −0.47∗∗∗ 0.89∗∗∗

(0.14) (0.12) (0.12) (0.12) (0.13) (0.13)Age −0.01 0.00 −0.04∗∗ 0.03 0.03 −0.05∗∗∗

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)Median Earnings x Low Income or First Gen. −0.04 −0.20 0.05 −0.09 −0.08 −0.02

(0.28) (0.26) (0.27) (0.26) (0.28) (0.28)Median Earnings x Low Income and First Gen. −0.27 −0.57 −0.05 −0.45 −0.54 0.14

(0.44) (0.39) (0.43) (0.41) (0.42) (0.45)R2 0.04 0.02 0.06 0.03 0.04 0.08F-Test p-value 0.79 0.32 0.98 0.49 0.41 0.94N 2005 2005 2005 2005 2005 2005∗∗∗p < 0.005, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table 10: OLS estimates of expected probability of completion per major on treatmentindicator and interaction between treatment indicator and family background status. Weexclude senior students from analysis. Regressions also include demographic and academiccontrols. Robust standard errors in parentheses. Three stars indicate statistical significanceat the level determined by the Bonferroni correction.

To help illustrate the interaction effects, we calculate the predicted average log probability

of completing a degree in each major by family background status and treatment status. We

also calculate the average difference in major completion probability between the treatment

groups. These results are presented in Table 11. Similar to the regression estimates, the

predicted difference in major choice probability is statistically indistinguishable from zero

for both SES groups and in all major fields.

26

Predicted CompletionMajor/Treatment Condition No Information Median Earnings Difference CIBusiness/High-SES -4.63 -4.67 -0.04 (-0.35, 0.28)Business/Low-SES -4.71 -5 -0.29 (-2.08, 0.56)Education/High-SES -5.18 -4.97 0.21 (-0.06, 0.50)Education/Low-SES -4.71 -5.04 -0.33 (-1.02, 0.41)Health/High-SES -4.51 -4.5 0.02 (-0.26, 0.31)Health/Low-SES -4.31 -4.31 -0.01 (-0.73, 0.8)Humanities/High-SES -5.08 -5.04 0.04 (-0.24, 0.33)Humanities/Low-SES -4.59 -4.98 -0.4 (-1.11, 0.37)Social Science/High-SES -4.35 -4.27 0.08 (-0.21, 0.38)Social Science/Low-SES -3.93 -4.38 -0.44 (-1.19, 0.37 )STEM/High-SES -3.79 -3.75 0.05 (-0.27, 0.33)STEM/Low-SES -3.83 -3.61 0.22 (-0.61, 1.05)

Table 11: Predicted mean expected log probability of degree completion, by treatment con-ditions and SES status. Column Low-SES shows predicted mean expected probability forlow-income and first-generation students; column High-SES shows predicted mean expectedprobability for neither low-income nor first-generation students. Column Difference showsthe estimated difference in expected probability between the two treatment groups. ColumnCI reports the 95% confidence interval on the estimated difference.

5.4. Limitations

Given these findings, we note several limitations to this analysis. First, our sample of

respondents is from a population already enrolled in college, such that our sample includes

respondents who have already selected into college, leaving out of the sample those who

did not select into college. The selection problem affects the results when those who have

selected into college are more informed about labor market outcomes than those who do not

go to college, perhaps due to resources used during the college search or provided by the

university itself.

Second, respondents who select into the survey may differ in observable and unobservable

characteristics from the university population overall. We show in Table 1 that the sample

is generally similar to the overall university population on all observables except gender.25

25Studies of response and non-response have noted that women are more likely than men to respond during

27

We cannot assess bias on unobservables. Those who respond, for example, may have

greater motivation to take the survey because they are interested in the topic of career

choice. This pattern would lead to bias when those who respond have already investigated

careers and labor market information, while those who do not respond have not sought out

this information.

One of our post-treatment survey questions reveals that the majority of students ex-

pressed that the labor market information presented in the survey was useful to them; sixty-

two percent of respondents listed the information as “useful” or “very useful,” while less

than seven percent listed the information as “useless” or “very useless.” To the extent that

respondents had already sought out labor market information, they may consider our in-

formation treatment as not being useful since they have already seen the data elsewhere.

These results, as well as responses to open-ended questions in our survey, suggest that many

respondents had either not seen the data before or, having seen it, still find use in seeing it

again.

Finally, our analysis of the treatment impact on low-income, first-generation status stu-

dents does not have a causal interpretation. Our experimental design only randomizes the

information provided to respondents; we cannot randomize family background. All estimates

of interaction effects that we present are associations, not causal effects.

6 Discussion

We ask three research questions in this paper: 1) Does providing students with labor mar-

ket information about median earnings change their own earnings expectations? 2) Does

providing students with labor market information about median earnings change their own

expectations about the field of study in which they intend to complete a degree, relative to

students who do not see any labor market information? 3) Does the effect of labor market

online surveys taken by the student body (Underwood et al. 2000).

28

information about median earnings vary systematically by student’s socioeconomic status?

We find that showing respondents median earnings alone reduces respondents’ own earn-

ings expectations. On average, students respond with expectations closer toward the national

median. Consistent with the idea of costly search (Hastings et al. 2015), we find that the

information treatment has a larger effect on respondents’ expectations in their counterfac-

tual majors. Though the information treatment has a substantively large impact on earnings

expectations, we find little evidence of a causal effect on major choice. Other researchers

have found a similar pattern (Kerr et al. 2014; Wiswall and Zafar 2015).

The descriptive findings for research question 3 are generally not consistent with a model

predicting that low-SES respondents hold more biased or inaccurate beliefs about labor

market information. Rather, high-SES respondents in the No Information condition seem to

overestimate the amount of money that graduates earn. In Business and Humanities, high-

SES respondents who see median earnings information have lower expectations than high-

SES respondents who see no labor market information. For Education, however, the pattern

is reversed: low-SES respondents who see no information have lower earnings expectations

than low-SES respondents who see the median earnings.

Disclosure of labor market information on college graduates has become an area of na-

tional focus for at least three reasons. Policymakers and universities are seeking to expand

access to higher education and ensure more equitable distribution of students across majors,

particularly in STEM fields. National policymakers have also prioritized reducing student

college debt, and one aspect of debt reduction is to encourage students to choose academic

fields such that they gain skills valued by the labor market. Understanding how information

affects the student choice process is crucial to the success of these efforts.

Our work builds on the policy relevance of scholars investigating the effect of publicly

available scorecards used to inform students about the returns to college and specific majors

(Huntington-Klein 2016a; Hurwitz and Smith 2016). Earnings information changes students’

29

expectations for their own future earnings, though the disclosure of earnings information

changes higher-SES respondents’ expectations more than low-SES respondents. However,

the evidence that earnings information ultimately changes students’ preferences over majors

is more limited. These findings therefore add to the body of literature that shows students

primarily weigh non-financial factors when making decisions about college major choice.

30

References

Ali, S. R., McWhirter, E. H., and Chronister, K. M. (2005). Self-efficacy and vocational

outcome expectations for adolescents of lower socioeconomic status: A pilot study. Journal

of Career Assessment, 13(1):40–58.

Arcidiacono, P., Hotz, J., and Kang, S. (2012). Modeling college major choice using elicited

measures of expectations and counterfactuals. Journal of Econometrics, 166(1):3–16.

Astin, A. (1993). What Matters in College? Four Critical Years Revisited. Jossey-Bass, San

Francisco.

Attanasio, O. P. and Kaufmann, K. M. (2012). Educational choices and returns on the labour

and marriage markets: Subjective expectations, gender and decision making. Unpublished

Manuscript.

Avery, C. (2009). The effects of college counseling on high-ability, low-income students:

Results of a pilot study with a randomized control trial. National Bureau of Economic

Research Working Paper Number 16359.

Avery, C. and Kane, T. J. (2004). Student Perceptions of College Opportunities: The Boston

COACH Program, pages 355–394. In College Choices: The Economics of Where to Go,

When to Go, and How to Pay for It. University of Chicago Press, Chicago.

Baker, R., Bettinger, E., Jacob, B., and Marinescu, I. (2017). The effect of labor market

information on community college students’ major choice. NBER Working Paper 23333.

Beattie, I. R. (2002). Are all adolescent econometricians created equal: Racial, class, and

gender differences in college enrollment. Sociology of Education, 75(1):19–43.

Berger, M. C. (1988). Predicted future earnings and choice of college major. Industrial and

Labor Relations Review, 41(3):418–429.

31

Bettinger, E. P., Long, B. T., Oreopoulos, P., and Sanbonmatsu, L. (2012). The role of

simplification and information in college decisions: Results from the H&R Block FAFSA

experiment. Quarterly Journal of Economics, 127(3):1205–1242.

Betts, J. R. (1996). What do students know about wages: Evidence from a survey of

undergraduates. Journal of Human Resources, 31(1):27–56.

Bleemer, Z. and Zafar, B. (2015). Intended college attendance: Evidence from an experiment

on college returns and costs. Federal Reserve Staff Report No. 739.

Botelho, A. and Pinto, L. C. (2004). Students’ expectations of the economic returns to

college education: Results of a controlled experiment. Economics of Education Review,

23(6):645–653.

Bourdieu, P. (1984). Distinction: A Social Critique of the Judgment of Taste. R. Nice

(Trans.). Harvard University Press, Cambridge.

Bourdieu, P. (1986). Forms of capital. In Richardson, J., editor, Handbook for Theory and

Research for the Sociology of Education. Greenwood Press.

Caner, A. and Okten, C. (2010). Risk and career choice: Evidence from Turkey. Economics

of Education Review, 29(6):1060–1075.

Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of

Sociology, 94:S95–S120.

Davies, S. and Guppy, N. (1997). Fields of study, college selectivity, and student inequalities

in higher education. Social Forces, 75(4):1417–1438.

Delany, L., Harmon, C., and Redmond, C. (2011). Parental education, grade attainment

and earnings expectations among university students. Economics of Education Review,

30:1136–1152.

32

Dominitz, J. and Manski, C. F. (1996). Eliciting student expectations of the returns to

schooling. Journal of Human Resources, 31(1):1–26.

Engle, J. and Tinto, V. (2008). Moving Beyond Access: College Success for Low-Income,

First-Generation Students. Washington, DC: Pell Institute for the Study of Opportunity

in Higher Education.

Fryer, Jr., R. G. (2013). Information and student achievement: Evidence from a cellular

phone experiment. Working Paper.

Goyette, K. A. and Mullen, A. L. (2006). Who studies the arts and sciences? Social back-

ground and the choice and consequences of undergraduate field of study. Journal of Higher

Education, 77(3):497–538.

Hastings, J., Neilson, C., Ramirez, A., and Zimmerman, S. (2015). (Un)informed college

and major choice: Evidence from linked survey and administrative data. National Bureau

of Economic Research Working Paper 21330.

Hastings, J., Neilson, C., and Zimmerman, S. (2017). The effects of earnings disclosure

on college enrollment decisions. National Bureau of Economic Research Working Paper

21300.

Horn, L. and Nunez, A.-M. (2000). Mapping the Road to College: First-Generation Students’

Math Track, Planning Strategies, and Context of Support. NCES 2000-153, Washington,

DC: National Center for Education Statistics.

Hoxby, C. and Turner, S. (2013). Expanding college opportunities for high-achieving, low-

income students. Stanford Institute for Economic Policy Research Report 12-014.

Hu, S. and Wolniak, G. (2010). Initial evidence of the influence of college student engagement

on early career earnings. Research in Higher Education, 51(8):750–766.

33

Hu, S. and Wolniak, G. C. (2013). College student engagement and early career earnings:

Differences by gender, race/ethnicity, and academic preparation. Review of Higher Edu-

cation, 36(2):211–213.

Huntington-Klein, N. (2016a). The search: The effect of the college scorecard on interest in

colleges. Working Paper.

Huntington-Klein, N. (2016b). ‘(Un)informed college and major choice’: Verification in an

alternate setting. Economics of Education Review, 53:159–163.

Hurwitz, M. and Smith, J. (2016). Student responsiveness to earnings data in the college

scorecard. Working Paper.

Imai, K., King, G., and Lau, O. (2008). Zelig: Everyone’s statistical software.

http://GKing.harvard.edu/zelig.

Ishitani, T. T. (2006). Studying attrition and degree completion behavior among students

in the United States. Journal of Higher Education, 77(5):861–885.

Jensen, R. (2010). The (perceived) returns to education and the demand for schooling. The

Quarterly Journal of Economics, 125(2):515–548.

Kelly, A. P. (2015). High costs, uncertain benefits. American Enterprise Institute

Center on Higher Education Reform. Available at www.aei.org/publication/

high-costs-uncertain-benefits/. Last accessed July 10 2017.

Kerr, S., Pekkarinen, T., Sarvimaki, M., and Uusitalo, R. (2014). Educational choice and

information on labor market prospects: A randomized field experiment. Working Paper.

Lareau, A. (1993). Home Advantage: Social Class and Parental Intervention in Elementary

Education. Palmer Press, New York.

34

Lareau, A. and Weininger, E. B. (2003). Cultural capital in educational research: A critical

assessment. Theory and Society, 32(5/6):567–606.

Lent, R. W., Brown, S. D., and Hackett, G. (2000). Contextual supports and barriers to

career choice: A social cognitive analysis. Journal of Counseling Psychology, 47(1):36–49.

Lundy-Wagner, V. C., Veenstra, C. P., Orr, M. K., and Ramirez, N. M. (2014). Gaining

access or losing ground?: Socioeconomic disadvantaged students in undergraduate engi-

neering 1994-2003. Journal of Higher Education, 85(3):339–369.

Ma, Y. (2009). Family socioeconomic status, parental involvement, and college major

choices—gender, race/ethnic, and nativity patterns. Sociological Perspectives, 52(2):211–

234.

Manski, C. F. (1993). Adolescent Econometricians: How Do Youth Infer the Returns to

Schooling? In Studies of Supply and Demand in Higher Education. Eds. Charles T.

Clotfelter and Michael Rothschild, Chapter 2. University of Chicago Press: Chicago.

Manski, C. F. and Wise, D. A. (1983). College Choice in America. Harvard University Press,

Boston.

McDonough, P. M. (1997). Choosing Colleges: How Social Class and Schools Structure

Opportunity. State University of New York Press, Albany.

McDonough, P. M. and Calderone, S. (2006). The meaning of money: Perceptual differences

between college counselors and low-income families about college costs and financial aid.

American Behavioral Scientist, 49(12):1703–1718.

Metheny, J. and McWhirter, E. H. (2013). Contributions of social status and family support

to college students’ career decision self-efficacy and outcome expectations. Journal of

Career Assessment, 21(3):378–394.

35

NCES (2012). Improving the measurement of socioeconomic status for the national assess-

ment of educational progress: A theoretical foundation. National Center for Education

Statistics. Available at nces.ed.gov/nationsreportcard/researchcenter/

socioeconomic_factors.aspx. Last accessed July 10 2017.

Nguyen, T. (2013). Information, role models and perceived returns to education: Experimen-

tal evidence from Madagascar. The World Bank. Word Bank Gender Impact Evaluation

Database. Washington, D.C.

Olson, L. and Rosenfeld, R. A. (1984). Parents and the process of gaining access to student

financial aid. Journal of Higher Education, 55(4):455–480.

Paulsen, M. (2001). The Economics of Human Capital and Investment in Higher Education,

pages 55–94. In The Finance of Higher Education: Theory, Research, Policy, and Practice.

Agathon Press, New York.

Perna, L. W. (2000). Differences in the decision to attend college among African Americans,

Hispanics, and Whites. Journal of Higher Education, 71(2):117–141.

Perna, L. W. (2006). Studying College Access and Choice: A Proposed Conceptual Model,

pages 99–157. In Higher Education: Handbook of Theory and Research, Vol XXI. Springer.

Perna, L. W. and Titus, M. A. (2005). The relationship between parental involvement as

social capital and college enrollment: An examination of racial/ethnic group differences.

Journal of Higher Education, 76(5):485–518.

Reuben, E., Wiswall, M. J., and Zafar, B. (2016). Preferences and biases in educational

choices and labour market expectations: Shrinking the black box of gender. Forthcoming,

Economic Journal.

36

Rouse, C. E. (2004). Low-income students and college attendance: An exploration of income

expectations. Social Science Quarterly, 85(5):1299–1317.

Toutkoushian, R. K. (2001). Do parental income and educational attainment affect the

initial choices of New Hampshire’s college-bound students? Economics of Education

Review, 20(3):245–262.

Trusty, J., Robinson, C. R., Plata, M., and Ng, K.-M. (2000). Effects of gender, socioe-

conomic status, and early academic performance on postsecondary educational choice.

Journal of Counseling & Development, 78(4):463–472.

Underwood, D., Kim, H., and Matier, M. (2000). To mail or to web: Comparisons of survey

response rates and respondent characteristics. Paper presented at the Annual Forum of

the Association for Institutional Research, Cincinnati, OH, May 21-24.

Walpole, M. B. (2003). Socioeconomic status and college: How SES affects college experi-

ences and outcomes. Review of Higher Education, 27(1):45–73.

Wells, R. S. and Lynch, C. M. (2012). Delayed college entry and the socioeconomic gap:

Examining the roles of student plans, family income, parental education, and parental

occupation. Journal of Higher Education, 83(5):671–697.

Willis, R. J. and Rosen, S. (1979). Education and self-selection. Journal of Political Economy,

87(5):S7–S36.

Wiswall, M. and Zafar, B. (2015). Determinants of college major choice: Identification using

an information experiment. Review of Economic Studies, 82(2):791–824.

Wooldridge, J. M. (2012). Introductory Econometrics. South-Western Cengage Learning,

Ohio, 5th edition.

37

A Supplemental Tables Not for Publication

In order to reduce the length of the main manuscript, we have placed several items in the

supplementary materials.

A.1. Drop Data Checks

In this section, we analyze differences between respondents who dropped out and those who

completed the survey. We focus on differences in several key variables that are likely related

to post-graduate labor market expectations: family background, gender, SAT scores, and

academic class level. In short, we only find differences between those who complete and those

who do not complete when analyzing SAT scores. Those who do not complete the survey

are significantly more likely to have a lower SAT score on the math and verbal components,

and more likely to have an SAT score missing from the dataset. This finding suggests that

those who did take the survey have, on average, higher academic performance on the SAT

A.1.1 Family Background

Complete Non CompleteFamily Background Percent PercentBase 79 21Low Income or First Gen. 79 21Low Income and First Gen. 76 24

Table A1: χ-squared = 2.5239, df = 2, p-value = 0.2831

A.1.2 Gender

Complete Non CompleteGener Percent PercentFemale 78 22Male 80 20

Table A2: χ-squared = 1.7338, df = 1, p-value = 0.1879

38

A.1.3 SAT

SAT Math SAT Verbalmean mean

Complete 608 579Non Complete 591 564

Table A3: ANOVA for Math SAT, F = 23, p < 0.01. ANOVA for Verbal SAT, F = 17,p < 0.01.

0 1 0 1Percent Percent Percent Percent

Complete 79 21 79 21Non Complete 74 26 74 26

Table A4: χ-squared = 11.525, df = 1, p − value < 0.01. Results the same for both mathand verbal SAT missingness.

A.1.4 Class Level

Complete Non CompletePercent Percent

First Year 22 20Junior 26 26Senior 33 32Sophomore 19 21

Table A5: χ-squared = 4.3586, df = 4, p-value = 0.36

A.1.5 Family Background

Keep DropFamily Background Percent PercentBase 5 95First-Gen or Low-Income 5 95First-Gen and Low-Income 5 95

39

Business Education Health Humanities Social Science STEMMedian Earnings −0.10∗∗∗ −0.02 −0.08∗∗∗ −0.06∗∗∗ −0.06∗∗∗ −0.08∗∗∗

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)

Low Income or First Gen. −0.07∗∗∗ −0.03 −0.03 −0.02 −0.01 −0.03(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)

Low Income and First Gen. −0.08∗∗ −0.04 −0.04 −0.04 −0.05 0.01(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)

Gender 0.06∗∗∗ 0.09∗∗∗ 0.03· 0.05∗∗ 0.02 0.03(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)

Age 0.00 0.01∗∗∗ −0.00 0.01∗∗∗ 0.01∗∗∗ −0.00(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)

Asian 0.00 −0.04 −0.02 −0.10∗∗ −0.18∗∗∗ −0.05(0.03) (0.04) (0.03) (0.04) (0.04) (0.03)

Latino −0.04 −0.05 −0.09∗ −0.09∗ −0.13∗∗∗ −0.06·

(0.04) (0.04) (0.04) (0.04) (0.04) (0.03)Unknown 0.05 −0.03 0.05 −0.08 −0.04 −0.01

(0.05) (0.07) (0.07) (0.07) (0.07) (0.06)Multiple Race −0.04 −0.11∗ −0.06 −0.14∗ −0.10· −0.04

(0.05) (0.05) (0.06) (0.06) (0.05) (0.05)Caucasian −0.02 −0.06· −0.08∗ −0.12∗∗∗ −0.15∗∗∗ −0.07∗

(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)Sophomore 0.00 −0.02 0.00 −0.05· −0.05· 0.02

(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)Junior 0.01 0.00 0.01 −0.04 −0.01 −0.01

(0.03) (0.03) (0.03) (0.03) (0.03) (0.02)Senior 0.04 −0.01 0.01 −0.04 −0.04 0.01

(0.03) (0.03) (0.03) (0.03) (0.03) (0.02)Somewhat Likely 0.01 0.02 0.04 0.03 0.03 0.02

(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)Somewhat Unlikely −0.01 −0.00 −0.04 −0.03 −0.03 −0.02

(0.03) (0.04) (0.04) (0.04) (0.04) (0.04)Very Likely 0.02 0.02 0.09∗∗∗ 0.01 0.05· 0.02

(0.02) (0.03) (0.03) (0.03) (0.03) (0.03)Very Unlikely 0.02 −0.01 0.06 0.01 0.02 0.05

(0.04) (0.04) (0.04) (0.05) (0.05) (0.04)Campus 1 0.10∗∗∗ 0.00 −0.01 0.00 −0.02 0.06∗

(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)Campus 2 0.18∗∗∗ 0.06· 0.04 0.06 0.05 0.06

(0.04) (0.04) (0.04) (0.04) (0.04) (0.04)Intercept 11.10∗∗∗ 10.61∗∗∗ 11.18∗∗∗ 10.60∗∗∗ 10.72∗∗∗ 11.31∗∗∗

(0.06) (0.07) (0.06) (0.07) (0.07) (0.07)R2 0.04 0.02 0.02 0.02 0.03 0.02N 2957 2957 2957 2957 2957 2957∗∗∗p < 0.0042, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table A6: OLS estimates of expected earnings per major on treatment indicator. Regres-sions also include family background indicators, and demographic and academic controls.Robust standard errors in parentheses. Three stars indicate statistical significance at thelevel determined by the Bonferroni correction.

40

Business Education Health Humanities Social Science STEMMedian Earnings −0.09∗∗∗ −0.04 −0.05 0.08 −0.03 −0.04·

(0.03) (0.06) (0.04) (0.06) (0.04) (0.02)

Low Income or First Gen. −0.05 −0.10 −0.01 0.03 −0.03 −0.04(0.03) (0.07) (0.05) (0.07) (0.04) (0.03)

Low Income and First Gen. −0.11∗ −0.29∗ −0.00 0.05 −0.04 −0.03(0.05) (0.11) (0.06) (0.08) (0.06) (0.04)

Gender 0.07∗∗ 0.07 0.11∗ 0.06 −0.03 0.07∗∗∗

(0.03) (0.08) (0.05) (0.06) (0.05) (0.02)Age −0.00 0.01 0.00 0.00 0.01 −0.01

(0.00) (0.01) (0.00) (0.00) (0.00) (0.01)Asian 0.02 −0.20 −0.02 −0.06 −0.07 −0.09∗

(0.06) (0.15) (0.07) (0.11) (0.09) (0.04)Latino −0.01 −0.09 −0.00 −0.09 −0.05 −0.09·

(0.06) (0.16) (0.07) (0.11) (0.07) (0.05)Unknown 0.04 0.05 −0.03 −0.06 0.06 0.07

(0.09) (0.30) (0.08) (0.17) (0.13) (0.07)Multiple Race −0.04 −0.15 0.00 −0.12 −0.03 −0.05

(0.10) (0.14) (0.11) (0.15) (0.11) (0.06)Caucasian 0.01 −0.04 −0.07 −0.05 0.02 −0.12∗∗

(0.05) (0.14) (0.06) (0.10) (0.07) (0.04)Sophomore 0.03 −0.02 −0.02 −0.03 −0.23∗∗∗ 0.05

(0.04) (0.08) (0.07) (0.11) (0.07) (0.03)Junior 0.01 0.07 −0.06 −0.06 −0.16∗ 0.01

(0.04) (0.09) (0.06) (0.10) (0.07) (0.03)Senior 0.06 0.03 −0.08 −0.14 −0.28∗∗∗ −0.01

(0.04) (0.10) (0.06) (0.10) (0.07) (0.03)Somewhat Likely 0.06· −0.08 0.00 0.09 0.09 0.00

(0.04) (0.11) (0.09) (0.09) (0.07) (0.03)Somewhat Unlikely −0.04 −0.19 −0.01 0.02 0.00 −0.01

(0.04) (0.17) (0.12) (0.11) (0.09) (0.05)Very Likely 0.07· 0.05 0.08 0.19∗ 0.13· 0.03

(0.04) (0.11) (0.09) (0.08) (0.07) (0.03)Very Unlikely 0.09 −0.14 −0.03 0.20 0.20 0.03

(0.06) (0.15) (0.14) (0.13) (0.12) (0.05)Campus 1 0.03 −0.08 0.01 0.04 −0.12∗ 0.01

(0.05) (0.09) (0.04) (0.09) (0.06) (0.05)Campus 2 0.11∗ 0.16 0.06 0.11 0.03 0.08

(0.05) (0.13) (0.07) (0.12) (0.07) (0.06)R2 0.06 0.12 0.04 0.05 0.08 0.04N 739 252 486 326 544 1073∗∗∗p < 0.0042, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table A7: OLS estimates of expected earnings per major on treatment indicator. Regres-sions also include family background indicators, and demographic and academic controls.Sample limited to responses about preferred major. Robust standard errors in parentheses.Three stars indicate statistical significance at the level determined by the Bonferroni methodcorrection.

41

A.2. Regression Tables

Business Education Health Humanities Social Science STEMMedian Earnings −0.10∗∗∗ −0.02 −0.09∗∗∗ −0.08∗∗∗ −0.06∗∗∗ −0.10∗∗∗

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)

Low Income or First Gen. −0.08∗∗∗ −0.02 −0.03 −0.03 −0.02 −0.03(0.02) (0.02) (0.02) (0.02) (0.02) (0.03)

Low Income and First Gen. −0.06· −0.01 −0.06 −0.05 −0.06 0.03(0.03) (0.03) (0.04) (0.04) (0.04) (0.04)

Gender 0.04· 0.09∗∗∗ 0.04∗ 0.06∗∗ 0.04∗ −0.02(0.02) (0.02) (0.02) (0.02) (0.02) (0.03)

Age 0.00 0.01∗∗∗ 0.00 0.01∗∗∗ 0.01∗∗∗ 0.00(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)

Asian −0.02 −0.02 −0.03 −0.10∗∗ −0.18∗∗∗ −0.03(0.04) (0.04) (0.04) (0.04) (0.04) (0.04)

Latino −0.06 −0.05 −0.10∗ −0.10∗ −0.16∗∗∗ −0.05(0.04) (0.04) (0.04) (0.04) (0.05) (0.05)

Unknown 0.04 −0.05 0.06 −0.09 −0.07 −0.06(0.06) (0.07) (0.08) (0.08) (0.08) (0.09)

Multiple Race −0.04 −0.11· −0.07 −0.15∗ −0.12∗ −0.03(0.06) (0.06) (0.06) (0.06) (0.06) (0.06)

Caucasian −0.03 −0.07· −0.08∗ −0.12∗∗∗ −0.21∗∗∗ −0.03(0.04) (0.04) (0.04) (0.04) (0.04) (0.04)

Sophomore −0.01 −0.03 0.02 −0.06· −0.04 0.02(0.03) (0.03) (0.03) (0.03) (0.03) (0.04)

Junior 0.01 −0.00 0.02 −0.04 −0.01 0.01(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)

Senior 0.04 −0.01 0.03 −0.04 −0.02 0.04(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)

Somewhat Likely −0.02 0.03 0.03 0.02 0.01 0.04(0.03) (0.03) (0.03) (0.03) (0.03) (0.04)

Somewhat Unlikely 0.01 0.01 −0.04 −0.04 −0.04 −0.03(0.04) (0.04) (0.04) (0.05) (0.05) (0.05)

Very Likely 0.02 0.01 0.07∗ −0.02 0.03 0.01(0.03) (0.03) (0.03) (0.03) (0.03) (0.04)

Very Unlikely −0.00 0.01 0.06 −0.02 −0.01 0.06(0.05) (0.04) (0.05) (0.05) (0.05) (0.05)

Campus 1 0.14∗∗∗ 0.01 0.02 −0.00 0.01 0.07∗

(0.04) (0.03) (0.03) (0.03) (0.03) (0.03)Campus 2 0.18∗∗∗ 0.05 0.06 0.05 0.05 0.05

(0.05) (0.04) (0.04) (0.04) (0.04) (0.04)R2 0.04 0.02 0.02 0.03 0.03 0.02N 2218 2705 2471 2631 2413 1884∗∗∗p < 0.0042, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table A8: OLS estimates of expected earnings per major on treatment indicator. Regres-sions also include family background indicators, and demographic and academic controls.Sample limited to responses about counterfactual majors. Robust standard errors in paren-theses. Three stars indicate statistical significance at the level determined by the Bonferronicorrection.

42

Business Education Health Humanities Social Science STEMMedian Earnings −0.08 0.11 0.03 −0.03 0.01 0.06

(0.13) (0.11) (0.12) (0.11) (0.12) (0.13)Low Income or First Gen. −0.06 0.07 0.17 0.04 0.28· 0.11

(0.15) (0.13) (0.14) (0.13) (0.14) (0.15)Low Income and First Gen. −0.21 0.19 0.20 0.26 0.14 0.06

(0.23) (0.21) (0.23) (0.22) (0.22) (0.24)Gender 0.43∗∗∗ −0.27∗ −0.57∗∗∗ −0.37∗∗∗ −0.47∗∗∗ 0.88∗∗∗

(0.14) (0.12) (0.12) (0.12) (0.13) (0.13)Age −0.01 0.00 −0.04∗∗ 0.03 0.03 −0.05∗∗∗

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)Asian 0.48∗ −0.61∗∗∗ 0.07 −0.62∗∗∗ −1.17∗∗∗ 0.62∗∗

(0.23) (0.21) (0.23) (0.21) (0.22) (0.23)Latino 0.15 0.04 −0.24 0.17 −0.47∗ 0.12

(0.24) (0.22) (0.23) (0.23) (0.24) (0.24)Unknown 0.34 −0.44 −0.15 0.54 −0.14 0.28

(0.42) (0.39) (0.42) (0.42) (0.42) (0.47)Multiple Race 0.01 −0.01 −0.37 0.07 −0.05 0.13

(0.37) (0.34) (0.37) (0.35) (0.37) (0.37)Caucasian −0.02 −0.28 −0.53∗ −0.40∗ −0.48∗ −0.31

(0.21) (0.20) (0.21) (0.20) (0.21) (0.22)Sophomore −0.12 −0.02 −0.34∗ −0.03 0.17 −0.70∗∗∗

(0.17) (0.15) (0.16) (0.15) (0.16) (0.17)Junior −0.23 −0.18 −0.28· −0.28· 0.06 −0.65∗∗∗

(0.16) (0.14) (0.16) (0.15) (0.15) (0.16)Somewhat Likely 0.22 0.14 0.19 0.06 0.00 −0.02

(0.22) (0.18) (0.19) (0.19) (0.20) (0.20)Somewhat Unlikely 0.13 −0.03 0.06 −0.04 −0.28 0.10

(0.29) (0.25) (0.25) (0.27) (0.27) (0.27)Very Likely −0.70∗∗∗ 0.00 0.63∗∗∗ −0.29 −0.26 0.27

(0.20) (0.17) (0.18) (0.18) (0.19) (0.19)Very Unlikely −0.26 0.17 0.03 0.07 −0.54· 0.56

(0.35) (0.29) (0.29) (0.31) (0.31) (0.34)Campus 1 −0.24 −0.33· −1.00∗∗∗ 0.30· 0.05 0.68∗∗∗

(0.20) (0.18) (0.20) (0.18) (0.19) (0.19)Campus 2 0.55∗ 0.08 −0.75∗∗∗ 0.36 0.32 0.08

(0.25) (0.23) (0.25) (0.23) (0.24) (0.24)R2 0.04 0.02 0.06 0.03 0.04 0.08N 2005 2005 2005 2005 2005 2005∗∗∗p < 0.005, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table A9: OLS estimates of expected probability of completion per major on treatmentindicator. Sample excludes senior respondents. Regressions also include family backgroundindicators, and demographic and academic controls. Robust standard errors in parenthe-ses. Three stars indicate statistical significance at the level determined by the Bonferronicorrection.

43

Business Education Health Humanities Social Science STEMIntercept 11.10∗∗∗ 10.61∗∗∗ 11.17∗∗∗ 10.60∗∗∗ 10.72∗∗∗ 11.31∗∗∗

(0.06) (0.07) (0.06) (0.07) (0.07) (0.07)Median Earnings −0.11∗∗∗ −0.04∗ −0.07∗∗∗ −0.08∗∗∗ −0.07∗∗∗ −0.08∗∗∗

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)Low Income or First Gen. −0.07∗ −0.04 −0.01 −0.05 −0.02 −0.02

(0.03) (0.03) (0.03) (0.03) (0.03) (0.02)Low Income and First Gen. −0.12∗∗∗ −0.10∗ −0.03 −0.09∗ −0.06 −0.01

(0.04) (0.04) (0.05) (0.04) (0.04) (0.04)Gender 0.06∗∗∗ 0.09∗∗∗ 0.03· 0.06∗∗ 0.02 0.03

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)Age 0.00 0.01∗∗∗ −0.00 0.01∗∗∗ 0.01∗∗∗ −0.00

(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)Asian 0.00 −0.04 −0.02 −0.11∗∗ −0.18∗∗∗ −0.05·

(0.03) (0.04) (0.03) (0.04) (0.04) (0.03)Latino −0.05 −0.05 −0.09∗ −0.09∗ −0.13∗∗∗ −0.06·

(0.04) (0.04) (0.04) (0.04) (0.04) (0.03)Unknown 0.05 −0.03 0.05 −0.08 −0.04 −0.01

(0.05) (0.07) (0.07) (0.07) (0.07) (0.06)Multiple Race −0.04 −0.11∗ −0.06 −0.14∗ −0.10· −0.04

(0.05) (0.05) (0.06) (0.06) (0.05) (0.05)Caucasian −0.02 −0.06· −0.08∗ −0.12∗∗∗ −0.15∗∗∗ −0.07∗

(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)Sophomore 0.00 −0.02 0.00 −0.05· −0.05· 0.02

(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)Junior 0.01 0.00 0.01 −0.04 −0.01 −0.01

(0.03) (0.03) (0.03) (0.03) (0.03) (0.02)Senior 0.04 −0.00 0.01 −0.04 −0.04 0.01

(0.03) (0.03) (0.03) (0.03) (0.03) (0.02)Somewhat Likely 0.00 0.02 0.04 0.03 0.03 0.02

(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)Somewhat Unlikely −0.01 −0.01 −0.04 −0.03 −0.03 −0.02

(0.03) (0.04) (0.04) (0.04) (0.04) (0.04)Very Likely 0.02 0.02 0.09∗∗∗ 0.01 0.05· 0.02

(0.02) (0.03) (0.03) (0.03) (0.03) (0.03)Very Unlikely 0.02 −0.01 0.06 0.01 0.02 0.05

(0.04) (0.04) (0.04) (0.05) (0.05) (0.04)Campus 1 0.10∗∗∗ 0.01 −0.01 0.01 −0.01 0.06∗

(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)Campus 2 0.18∗∗∗ 0.07· 0.04 0.06 0.05 0.06

(0.04) (0.04) (0.04) (0.04) (0.04) (0.04)Median Earnings x Low Income or First Gen. −0.01 0.03 −0.04 0.05 0.02 −0.01

(0.04) (0.04) (0.04) (0.04) (0.04) (0.04)Median Earnings x Low Income and First Gen. 0.09 0.12∗ −0.04 0.11· 0.02 0.05

(0.06) (0.06) (0.06) (0.06) (0.06) (0.05)R2 0.04 0.02 0.02 0.02 0.03 0.02F-Test p-value 0.22 0.10 0.59 0.17 0.90 0.55N 2957 2957 2957 2957 2957 2957∗∗∗p < 0.0042, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table A10: OLS estimates of expected earnings per major on treatment indicator, familybackground indicators, interactions between treatment and family background indicators,and demographic and academic controls, for respondents in their counterfactual major. Ro-bust standard errors in parentheses. Three stars indicate statistical significance at the leveldetermined by the Bonferroni method correction.

44

Business Education Health Humanities Social Science STEMIntercept −3.14∗∗∗ −4.43∗∗∗ −2.44∗∗∗ −4.95∗∗∗ −4.33∗∗∗ −2.77∗∗∗

(0.53) (0.46) (0.48) (0.48) (0.52) (0.50)Median Earnings −0.04 0.22 0.02 0.04 0.08 0.05

(0.16) (0.14) (0.15) (0.14) (0.15) (0.16)Low Income or First Gen. −0.04 0.16 0.14 0.09 0.32 0.12

(0.21) (0.19) (0.20) (0.18) (0.20) (0.21)Low Income and First Gen. −0.07 0.48· 0.23 0.50 0.42 −0.01

(0.32) (0.29) (0.31) (0.31) (0.32) (0.33)Gender 0.43∗∗∗ −0.27∗ −0.57∗∗∗ −0.38∗∗∗ −0.47∗∗∗ 0.89∗∗∗

(0.14) (0.12) (0.12) (0.12) (0.13) (0.13)Age −0.01 0.00 −0.04∗∗ 0.03 0.03 −0.05∗∗∗

(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)Asian 0.48∗ −0.60∗∗∗ 0.07 −0.62∗∗∗ −1.16∗∗∗ 0.62∗∗

(0.23) (0.21) (0.23) (0.21) (0.22) (0.23)Latino 0.15 0.04 −0.24 0.17 −0.47∗ 0.12

(0.24) (0.22) (0.23) (0.23) (0.24) (0.24)Unknown 0.34 −0.45 −0.15 0.53 −0.15 0.28

(0.42) (0.39) (0.42) (0.42) (0.42) (0.47)Multiple Race 0.01 −0.01 −0.37 0.08 −0.05 0.13

(0.37) (0.34) (0.37) (0.35) (0.37) (0.37)Caucasian −0.02 −0.27 −0.53∗ −0.39· −0.48∗ −0.31

(0.21) (0.20) (0.21) (0.20) (0.21) (0.22)Sophomore −0.12 −0.02 −0.34∗ −0.03 0.17 −0.70∗∗∗

(0.17) (0.15) (0.16) (0.15) (0.16) (0.17)Junior −0.23 −0.19 −0.28· −0.28· 0.05 −0.65∗∗∗

(0.16) (0.14) (0.16) (0.15) (0.16) (0.16)Somewhat Likely 0.22 0.14 0.19 0.06 0.00 −0.02

(0.22) (0.18) (0.19) (0.19) (0.20) (0.20)Somewhat Unlikely 0.13 −0.03 0.06 −0.04 −0.28 0.10

(0.29) (0.25) (0.25) (0.27) (0.27) (0.27)Very Likely −0.70∗∗∗ 0.00 0.63∗∗∗ −0.29 −0.26 0.27

(0.20) (0.17) (0.18) (0.18) (0.19) (0.19)Very Unlikely −0.26 0.17 0.03 0.07 −0.54· 0.56

(0.35) (0.29) (0.29) (0.31) (0.31) (0.34)Campus 1 −0.25 −0.34· −1.00∗∗∗ 0.29· 0.05 0.69∗∗∗

(0.20) (0.18) (0.20) (0.18) (0.19) (0.19)Campus 2 0.55∗ 0.07 −0.75∗∗∗ 0.35 0.30 0.08

(0.25) (0.23) (0.25) (0.23) (0.24) (0.24)Median Earnings x Low Income or First Gen. −0.04 −0.20 0.05 −0.09 −0.08 −0.02

(0.28) (0.26) (0.27) (0.26) (0.28) (0.28)Median Earnings x Low Income and First Gen. −0.27 −0.57 −0.05 −0.45 −0.54 0.14

(0.44) (0.39) (0.43) (0.41) (0.42) (0.45)R2 0.04 0.02 0.06 0.03 0.04 0.08F-Test p-value 0.79 0.32 0.98 0.49 0.41 0.94N 2005 2005 2005 2005 2005 2005∗∗∗p < 0.005, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1

Table A11: OLS estimates of expected probability of completion per major on treatmentindicator and interaction between treatment indicator and family background status. Weexclude senior students from analysis. Regressions also include demographic and academiccontrols. Robust standard errors in parentheses. Three stars indicate statistical significanceat the level determined by the Bonferroni correction.

45